Problem: rpl_switch_stm_row_mixed did not wait until row events generated by
INSERT DELAYED were written to the master binlog before it synchronized slave
with master. This caused sporadic errors where these rows were missing on
slave.
Fix: wait until all rows appear on the slave.
This is a backport, applying the same to 5.1-bugteam as was previously
applied to 6.0-rpl
On a slow environment like valgrind the test is vulnerable
because it does not check if slave has stopped at time
of the new session is requested `start slave;' -- disabling
test till it is fixed.
The test is vulnerable because it does not check if slave has stopped at time
of the new session is requested `start slave;'
Fixed with deploying explicitly wait_for_slave_to_stop synchronization macro.
Problem: If INSERT is immediately followed by SELECT in another thread,
the newly inserted rows may not be returned by the SELECT statement, if
ENGINE=myisam and @@concurrent_insert=1. This caused sporadic errors in
rpl_insert_id.
Fix: The test now uses ENGINE=$engine_type when creating tables (so that
innodb is used). It also turns off @@concurrent_insert around the critical
place, so that it works if someone in the future writes a test that sets
$engine_type=myisam before sourcing extra/rpl_tests/rpl_insert_id.test.
It also adds ORDER BY to all SELECTs so that the result is deterministic.
When flushing tables, there were a slight chance that the flush was occuring
between processing of two table map events. Since the tables are opened
one by one, it might result in that the tables were not valid and that sub-
sequent locking of tables would cause the slave to crash.
The problem is solved by opening and locking all tables at once using
simple_open_n_lock_tables(). Also, the patch contain a change to open_tables()
so that pre-locking only takes place when the trg_event_map is not zero, which
was not the case before (this caused the lock to be placed in thd->locked_tables
instead of thd->lock since the assumption was that triggers would be called
later and therefore the tables should be pre-locked).
Temporarily checking in an incorrect test case. Rationale: the impact of
this bug is negligible (it's almost a feature request). We need 5.1 to be
stable, and making a real fix is a bit risky. So the fix is postponed
to 6.0.
The test suite/rpl/t/rpl_innodb_bug28430.test was disabled because of
BUG#32247, but not re-enabled when BUG#32247 was fixed. I've re-enabled
it. The test and result file needed to be updated too.
Select of the test could not perform deterministically, because the table remains to be
updatable by the running event handler.
Fixed with changing verification to use a logical values instead of comparison
with a pre-recorded results.
Among two claimed artifacts the critical one is in that the Table map of
a query following the failing with a duplicate key error CREATE-SELECT is skipped from
instantionating (and thus binlogging). That leads to sending a "chopped" group of the data
row-events without the table map head to the slave.
The slave can not apply the only data row events.
It's not easy to force the slave to react with an error in such a case (the second complaint
on the bug report), because the lack of a table Rows_log_event::do_apply_event the data row event
handler is a common situation which normally designates the event has to be filtered out
basing on the repliation do/ingore rules decision.
Fixed: table map creating and binlogging is restored via deploying the standard cleanup call in
select_create::abort().
No error is reported if by chance the table map was not been binlogged.
Leaving this out to resolve with considering how to combine the do/ingore rules with the situation
when erronoulsy the Table_map is not written to binlog.
Disabled 'rpl_redirect', failure is sporadic and and the test is superfluous
rpl_packet.test, rpl_packet.result:
Removing race conditions from rpl_packet causing test to fail
If a binlog file is manually replaced with a namesake directory the internal purging did
not handle the error of deleting the file so that eventually
a post-execution guards fires an assert.
Fixed with reusing a snippet of code for bug@18199 to tolerate lack of the file but no other error
at an attempt to delete it.
The same applied to the index file deletion.
The cset carries pieces of manual merging.
improving a test that shows a failure.
the wait condition was for data in tables but the
log positions are updates after the data are unlocked.
So there was a time window
[after_table_unlock_for_select, log_pos_updated] where the
orig cond was true but log position might be changed.
the correct one is to expect the last pos of the
slave's insert in the output of show_slave_status on the
master.
The bug allow multiple executing transactions working with non-transactional
to interfere with each others by interleaving the events of different trans-
actions.
Bug is fixed by writing non-transactional events to the transaction cache and
flushing the cache to the binary log at statement commit. To mimic the behavior
of normal statement-based replication, we flush the transaction cache in row-
based mode when there is no committed statements in the transaction cache,
which means we are committing the first one. This means that it will be written
to the binary log as a "mini-transaction" with just the rows for the statement.
Note that the changes here does not take effect when building the server with
HAVE_TRANSACTIONS set to false, but it is not clear if this was possible before
this patch either.
For row-based logging, we also have that when AUTOCOMMIT=1, the code now always
generates a BEGIN/COMMIT pair for single statements, or BEGIN/ROLLBACK pair in the
case of non-transactional changes in a statement that was rolled back. Note that
for the case where changes to a non-transactional table causes a rollback due
to error, the statement will now be logged with a BEGIN/ROLLBACK pair, even
though some changes has been committed to the non-transactional table.
